Curated AI Learning Bookmarks
This page turns the WhatsApp bookmark export into a learning queue. The export contained 249 URLs and 245 exact-unique URLs. The useful links are organized here by when they become relevant; duplicates, advertisements, unrelated pages, opaque short links, and private documents are not part of the public curriculum.
Return to the main curriculum for the ordered syllabus and the current 30-day book-and-paper sprint.
Collection rule
A bookmark is not an assignment. Keep at most one primary course, one reference book, and one build project active. Everything else waits on its shelf until the current curriculum phase calls for it.
Status key
- Now — directly supports the current 30-day sprint
- Next — use after the sprint, during Foundations or Core ML
- Branch — open only after choosing the matching specialization
- Reference — consult for a specific question; do not read linearly
- Recheck — time-sensitive opportunity whose current status must be verified
Planned sequence
Now — support the 30-day sprint
Use these only when the corresponding book chapter needs another explanation.
- Caltech Learning From Data lectures — companion lectures for the primary book
- Sebastian Raschka’s machine-learning overview notes — compact orientation before deeper theory
- Chain rule, Jacobians, autograd, and tensor shapes — use when gradient shapes become confusing
- ProbabilityCourse.com — targeted probability revision
- MIT Fundamentals of Probability lecture notes — rigorous backup reference
- Mathematics for Machine Learning notes — Marc Peter Deisenroth — bridge between linear algebra, probability, and ML
- Mathematics for Machine Learning — Fleuret notes — concise reference for notation and core methods
- Representation Learning: A Review and New Perspectives — read after the sprint’s linear-model chapter
Exit condition: finish the sprint outputs. Do not add a second full course during this month.
Next — algorithms, software, and data foundations
Pick one algorithms course and one programming/data resource.
Algorithms and data structures
- UC Berkeley CS 170 — primary algorithms course
- Jeff Erickson’s Algorithms — free textbook and reference
- Open Data Structures — implementation-oriented text
- CSES Data Structures and Algorithms — structured problem practice
- Algorithms collection — supplementary implementations
- Maths, CS, and AI Compendium — map and reference, not a second curriculum
Programming and data systems
- Software Carpentry lessons — shell, Git, Python, and reproducible research basics
- Data Analysis with Python — practical Python/data work
- Big Data Platforms — distributed-data concepts
- Intro to Hadoop and MapReduce — historical context; skim after modern data-pipeline basics
- Pseudocode notes — communicating algorithms clearly
Mathematical support
- Cornell: determinants and row operations
- Cornell: real eigenvalues
- Cornell: invertible matrices
- NASA introduction to tensors — later reference, not an immediate prerequisite
- Stanford CS 205A numerical methods notes
- Stanford CS 205L lectures — numerical linear algebra and scientific computing
Exit condition: implement representative data structures, analyze their complexity, and complete a reproducible data-analysis project.
Next — core machine learning and deep learning
Choose one primary course. Treat the rest as alternative explanations.
Primary-course candidates
- Andrew Ng’s Machine Learning course — accessible first pass
- NYU Deep Learning — modern lecture sequence
- UBC Lectures on Machine Learning — theory and optimization lectures
- Stanford Engineering Everywhere — archive of complete engineering courses
- Foundations of AI/ML Course Atlas — compare coverage; do not make it another active course
Books and theory references
- Foundations of Machine Learning resources
- Machine Learning — Tom Mitchell
- Neural Networks and Deep Learning
- The Principles of Deep Learning Theory — advanced theory reference
- Mathematical Foundations of Deep Learning — use after core probability and optimization
- Convex Optimization lecture notes — use alongside the optimization phase
- Machine-learning teaching tricks — practical intuition and habits
- PaperCode — paper-to-implementation practice
Exit condition: compare several models on one dataset, include a baseline and ablation, and explain the dominant source of error.
Branch shelves
Open one shelf only after completing the common foundations.
Language models, transformers, and agents
Conceptual path
- Foundations of Large Language Models
- Stanford CS 324: Large Language Models
- Understanding Transformers and Attention Mechanisms — mathematical introduction
- Transformer math
- Understanding transformers beyond the math
- Interactive transformer explanation
- Embedding visualization
- Alisa’s Book of LLMs — broad reference
- Rush NLP — NLP course material and implementations
Build from the components
- Build and Train an LLM with JAX
- How to Train Your GPT
- OLMo Core — production-quality training code
- GPT-2 implementation — Laz4rz
- GPT-2 implementation — AnirudhhRamesh
- Implementing byte-pair encoding
- Another BPE walkthrough
- KV caching and attention
- SnapKV interactive walkthrough
- Cohere RAG cookbooks
- MicroThink — inspect after building a basic agent loop
- Generative Agents — memory, reflection, and planning
- Composer 2 technical report — coding-agent systems reference
Native multimodal models
Computer vision, representation, and generative models
- HPI Computer Vision — primary course candidate
- 3D Geometric Vision
- CLIP paper
- Contrastive representation learning
- TensorFlow Embedding Projector
- FiftyOne nuScenes explorer
- CLIP zero-shot classification notebook
- CLIP experiments notebook
- Dataset curation with CLIP and LeNet
- Layer-wise relevance propagation tutorial
- VQGAN-CLIP
- Diffusion Models from Scratch
- MIT diffusion lecture notes
- VideoFlexTok
- Understanding UMAP
- BERTopic
- Embedding Wikipedia archives
- Unsplash Lite 5K dataset
Reinforcement learning and robotics
Learn
- Deep Reinforcement Learning notes
- Robot Learning: A Tutorial
- ETH Robot Learning
- Princeton Introduction to Robotics
- Foundations of Robotics
- Building Intelligent Autonomous Navigation Agents
- Playing FPS Games with Deep Reinforcement Learning
- OpenAI Evolution Strategies
Build and optimize
- Hands-on Modern RL
- CS 5180 Reinforcement Learning
- Making RL fast
- MaxRL
- Dota RL bot
- Vision-language-action build notes
Mechanistic interpretability and alignment
Begin with the course hub, then reproduce one small result before reading the entire shelf.
- Mechanistic Interpretability — starting hub
- AI Alignment course
- A Mathematical Framework for Transformer Circuits
- Transformers Learn Shortcuts to Automata
- Transformer Grammars
- Interpreting Othello-GPT
- Progress measures for grokking
- A Toy Model of Universality
- Toy Models of Superposition: double descent
- Discovering Latent Knowledge Without Supervision
- Interpreting Neural Networks through the Polytope Lens
- Finding Neurons in a Haystack
- AttentionViz
- Steering GPT-2 XL
- Q-Probe
- Situational Awareness — argument to critique, not settled curriculum content
Computational neuroscience and active inference
- Neuromatch Computational Neuroscience
- Computational Neuroscience prerequisites
- Neuromatch Deep Learning prerequisites
- Neuroscience and artificial neural networks
- Brain input/output techniques
- Neuroplasticity
- Active inference textbooks
ML systems, hardware, and scaling
- Machine Learning at Scale
- GPU MODE
- Inside the M4 Apple Neural Engine
- Building a TPU from scratch
- Inside the chips
- NVIDIA Deep Learning Summer Camp
- Meta’s WebRTC modernization — software-systems case study
Research paper queue
Read these through the paper-card method on the main curriculum page. The order moves from broad foundations toward narrower or newer work.
Foundations and model behavior
- Representation Learning: A Review and New Perspectives
- Recurrent Neural Network Regularization
- The Principles of Deep Learning Theory
- Foundations of Large Language Models
- Mathematical Foundations of Deep Learning
Architectures and applications
Interpretability
- Transformer Grammars
- Transformers Learn Shortcuts to Automata
- Interpreting Neural Networks through the Polytope Lens
- Discovering Latent Knowledge Without Supervision
- Progress Measures for Grokking
- A Toy Model of Universality
- Finding Neurons in a Haystack
- AttentionViz
Robotics, control, and safety
- Playing FPS Games with Deep Reinforcement Learning
- Building Intelligent Autonomous Navigation Agents
- Robot Learning: A Tutorial
- Large-scale online deanonymization with LLMs — privacy and misuse case study
Project and evaluation tools
- Haystack hallucination score calculator
- TensorFlow Embedding Projector
- BERTopic
- NVIDIA topic-modeling blueprint
- Keras Molecules
- Learning Invariances
- Shap-Editor
- YoloGemma
- ActGPT
- BacKGProp
- NeurIPS competition hub — Recheck dates and eligibility
Career, community, and building habits
Build and communicate
- How to ship
- Learning how to learn
- Advice to young people
- Personal best
- Motivation and optimization
- How to make a font — example of a polished technical explainer
Job-search and interview references
Opportunities — recheck before acting
- Google Summer of Code — Recheck the current application cycle
- Hugging Face Builders — Recheck cohort status
- Agent Community — Recheck membership details
- EleutherAI SOAR — Recheck applications and eligibility
- ALPI Lab open positions — Recheck availability
Portfolio and bookshelf inspiration
These are references for presentation, research taste, and further reading—not curriculum requirements.
- Georgi Gerganov
- Laura Chen
- Raj Working
- Thomas Fel
- Yuntian Deng
- Daphne Demekas
- Misha Laskin
- HuskyDoge’s bookshelf
- Poojan Shah’s book collection
- ML and statistics book reviews
- Good reads
- Ilya Sutskever recommended reading
Import decisions
The following remain only in the source export:
- Private or access-token-bearing Google Docs, Drive folders, Forms, NotebookLM notebooks, Claude artifacts, and personal Notion pages
- Unlabeled YouTube videos, X posts,
t.coshort links, and redirect wrappers whose destination or purpose is unclear - Expired sales, past hackathons, dated event registrations, and promotional messages
- Duplicate URLs and duplicate versions of the same HPI course
- Copyright mirrors and unofficial book-file repositories when an official source is preferable
- Unrelated bookmarks such as restaurants and general-interest pages
When importing new bookmarks, record title, purpose, curriculum phase, and the next concrete action. A URL without those fields stays in the inbox rather than entering the curriculum.